Abstract: For problems of slow convergence speed and low precision existing in gear box fault diagnosis method based on traditional BP neural network, a gearbox fault diagnosis model based on Elman neural network was proposed. In the model, feature vectors are taken as input information and fault types as output information. An improved genetic algorithm is used to optimize weights and thresholds of Elman neural network, and the optimized Elman neural network is used for gear box fault diagnosis. The simulation results show that the model accelerates network convergence speed and improves accuracy and precision of gear box fault diagnosis.